Functional maturation in preterm infants measured by serial recording of cortical activity

Abstract

Minimally invasive, automated cot-side tools for monitoring early neurological development can be used to guide individual treatment and benchmark novel interventional studies. We develop an automated estimate of the EEG maturational age (EMA) for application to serial recordings in preterm infants. The EMA estimate was based on a combination of 23 computational features estimated from both the full EEG recording and a period of low EEG activity (46 features in total). The combination function (support vector regression) was trained using 101 serial EEG recordings from 39 preterm infants with a gestational age less than 28 weeks and normal neurodevelopmental outcome at 12 months of age. EEG recordings were performed from 24 to 38 weeks post-menstrual age (PMA). The correlation between the EMA and the clinically determined PMA at the time of EEG recording was 0.936 (95%CI: 0.932-0.976; n = 39). All infants had an increase in EMA between the first and last EEG recording and 57/62 (92%) of repeated measures within an infant had an increasing EMA with PMA of EEG recording. The EMA is a surrogate measure of age that can accurately determine brain maturation in preterm infants.

Conflict of interest statement

The authors declare that they have no competing interests.

Figures

Figure 1

The change in EEG maturational age (EMA) with the post-menstrual age of EEG recording (PMA) in individual infants with normal neurodevelopmental outcome at 12 months of age. (A) Infants with constantly increasing EMA with increasing PMA (normal growth) (n = 29). (B) Infants with instances of decreasing EMA with increasing PMA (deviant growth) (n = 5). Note that, these instances of deviant growth were due to a single outlier EMA (EMA from a recording with an absolute difference between EMA and PMA greater than 2 weeks). Figures show an EMA estimate based on 46 features with artefact detection and are calculated within a leave-one-infant-out cross-validation on a per recording basis (n = 101). The numbers denote infants and the additional lines track changes in longitudinal recordings: green lines denote an increase in EMA with increasing PMA, while red lines denote a decrease in EMA between consecutive EEG recording despite an increasing PMA. Underlying black lines indicate 0 (solid), 1 (dashed) and 2 (dashed) week differences between the EMA and the PMA.

Figure 2

The result of feature selection during the development of the EEG maturational age. The number of times a feature was selected during cross-validation is overlaid with the correlation of each feature with the PMA (cross-validation iterations = 39, feature number = 46); feature numbers align with feature labels given in Fig. 3A. Note that, the features selected most often are not always the features with the highest individual correlation between EMA and PMA. The legend refers to the segment of EEG each feature was estimated on: full denotes full 1 h EEG epoch and low SAT% refers to a segment of EEG with low SAT activity. Similarly, feature correlations are denoted by ‘ + ’ when estimated on the full 1 h epoch and ‘x’ when estimated on the low SAT% segment of the epoch.

Figure 3

EEG maturational age (EMA) measurement in preterm infants. (A) The flow diagram of the EMA algorithm. Percentages in brackets after features refer to the percentile estimated by the feature, for example, envelope (50%) is the 50th percentile or median envelope calculated in the period of interest. SAT is spontaneous activity transient, rEEG is the range EEG, and RMS is the root-mean-square. In this case, delta was 0–3 Hz, theta was 3–8 Hz, alpha was 8–15 Hz and beta was 15–30 Hz. (B) The EEG dataset used in this study. The figure shows the distribution of post-menstrual age (PMA) of EEG recordings. The table shows additional demographics of the EEG cohort used in this study - data are summarised as median (interquartile range), except gender which is given as a count. Cohort sample size was 43 infants, 152 EEG recordings and 1080, one hour EEG epochs. (C) Evaluating the EMA within a leave-one-subject cross-validation. The dataset was iteratively split into training and test sets and the efficacy of the EMA was assessed by comparing the EMA from the test set to the PMA of EEG recording. The diagonal lines denote errors of plus and minus 0, 1 and 2 weeks.